if(!require("EBImage")){
  install.packages("BiocManager")
  BiocManager::install("EBImage")
}
if(!require("R.matlab")){
  install.packages("R.matlab")
}
if(!require("readxl")){
  install.packages("readxl")
}

if(!require("dplyr")){
  install.packages("dplyr")
}
if(!require("readxl")){
  install.packages("readxl")
}

if(!require("ggplot2")){
  install.packages("ggplot2")
}

if(!require("caret")){
  install.packages("caret")
}

if(!require("glmnet")){
  install.packages("glmnet")
}

if(!require("WeightedROC")){
  install.packages("WeightedROC")
}

if(!require("gbm")){
  install.packages("gbm")
}

if(!require("tidyverse")){
  install.packages("tidyverse")
}

library(R.matlab)
library(readxl)
library(dplyr)
library(EBImage)
library(ggplot2)
library(caret)
library(glmnet)
library(WeightedROC)
library(gbm)
library(tidyverse)

Step 0 set work directories

set.seed(2020)
setwd(getwd())

Provide directories for training images. Training images and Training fiducial points will be in different subfolders.

train_dir <- "../data/train_set/" # This will be modified for different data sets.
train_image_dir <- paste(train_dir, "images/", sep="")
train_pt_dir <- paste(train_dir,  "points/", sep="")
train_label_path <- paste(train_dir, "label.csv", sep="")

Step 1: set up controls for evaluation experiments.

In this chunk, we have a set of controls for the evaluation experiments.

run.cv <- TRUE # run cross-validation on the training set
sample.reweight <- TRUE # run sample reweighting in model training
K <- 5  # number of CV folds
run.feature.train <- TRUE # process features for training set
run.test <- TRUE # run evaluation on an independent test set
run.feature.test <- TRUE # process features for test set

Using cross-validation or independent test set evaluation, we compare the performance of models with different specifications.

In this part, we tune parameters n.trees (total number of trees) and shrinkage (learning rate) for boosted decision stumps (Gradient Boosting Machine).

#GBM parameters
n.trees  <- c(40,50,60,70) #40-70
shrinkage <- c(0.01,0.05,0.1,0.15) #0.01-1

#model_labels = paste("GBM with n.trees =", n.trees,"and shrinkage =",shrinkage)

Step 2: import data and train-test split

#train-test split
info <- read.csv(train_label_path)
n <- nrow(info)
n_train <- round(n*(4/5), 0)
train_idx <- sample(info$Index, n_train, replace = F) 
test_idx <- setdiff(info$Index, train_idx)

If you choose to extract features from images, such as using Gabor filter, R memory will exhaust all images are read together. The solution is to repeat reading a smaller batch(e.g 100) and process them.

n_files <- length(list.files(train_image_dir))

image_list <- list()
for(i in 1:100){
   image_list[[i]] <- readImage(paste0(train_image_dir, sprintf("%04d", i), ".jpg"))
}

Fiducial points are stored in matlab format. In this step, we read them and store them in a list.

#function to read fiducial points
#input: index
#output: matrix of fiducial points corresponding to the index
readMat.matrix <- function(index){
     return(round(readMat(paste0(train_pt_dir, sprintf("%04d", index), ".mat"))[[1]],0))
}

#load fiducial points
fiducial_pt_list <- lapply(1:n_files, readMat.matrix) 

save(fiducial_pt_list, file="../output/fiducial_pt_list.RData")

Step 3: construct features and responses

Figure1

Figure1

feature.R should be the wrapper for all your feature engineering functions and options. The function feature( ) should have options that correspond to different scenarios for your project and produces an R object that contains features and responses that are required by all the models you are going to evaluate later.

source("../lib/feature.R")
tm_feature_train <- NA
if(run.feature.train){
  tm_feature_train <- system.time(dat_train <- feature(fiducial_pt_list, train_idx))
  save(dat_train, file="../output/feature_train.RData")
}else{
  load(file="../output/feature_train.RData")
}

tm_feature_test <- NA
if(run.feature.test){
  tm_feature_test <- system.time(dat_test <- feature(fiducial_pt_list, test_idx))
  save(dat_test, file="../output/feature_test.RData")
}else{
  load(file="../output/feature_test.RData")
}

Step 4: Train a classification model with training features and responses

Call the train model and test model from library.

train.R and test.R should be wrappers for all your model training steps and your classification/prediction steps.

source("../lib/train_baseline.R") 
source("../lib/test_baseline.R")
source("../lib/cv_baseline.R")

Model selection with cross-validation

  • Do model selection by choosing among different values of training model parameters.
#feature_train = dat_train[, -6007]
#label_train = dat_train$label
feature_train = as.matrix(dat_train[, -6007]) 
label_train = as.integer(dat_train$label)
if(run.cv){
  
mean.error_cv <- matrix(0, nrow = length(n.trees), ncol=length(shrinkage))
sd.error_cv <- matrix(0, nrow = length(n.trees), ncol=length(shrinkage))
mean.AUC_cv <- matrix(0, nrow = length(n.trees), ncol=length(shrinkage))
sd.AUC_cv <- matrix(0, nrow = length(n.trees), ncol=length(shrinkage))

  for(i in 1:length(n.trees)){
    cat("n.trees = ", n.trees[i], "\n")
    
    for (j in 1:length(shrinkage)){
    cat("shrinkage = ", shrinkage[j], "\n")
      
      res_cv<- cv.function(feature_train, 
                              label_train, 
                              K, 
                              n.trees[i],
                              shrinkage[j],
                           reweight = sample.reweight)
      
      mean.error_cv[i,j]<-res_cv[1]
      sd.error_cv[i,j]<-res_cv[2]
      mean.AUC_cv[i,j]<-res_cv[3]
      sd.AUC_cv[i,j]<-res_cv[4]
    }
  }

save(mean.error_cv, file="../output/mean.error_cv.RData")
save(sd.error_cv, file="../output/sd.error_cv.RData")
save(mean.AUC_cv, file="../output/mean.AUC_cv.RData")
save(sd.AUC_cv, file="../output/sd.AUC_cv.RData")

}else{
  load("../output/mean.error_cv.RData")
  load("../output/sd.error_cv.RData")
  load("../output/mean.AUC_cv.RData")
  load("../output/sd.AUC_cv.RData")
  
  }

Visualize cross-validation results.

df.mean.error<-data.frame(mean.error_cv)%>%
  setNames(shrinkage)%>%
  mutate(n.trees=n.trees)%>%
  pivot_longer(c(-n.trees),names_to = "shrinkage",values_to = "mean.error")

df.sd.error<-data.frame(sd.error_cv)%>%
  setNames(shrinkage)%>%
  mutate(n.trees=n.trees)%>%
  pivot_longer(c(-n.trees),names_to = "shrinkage",values_to = "sd.error")

df.mean.AUC<-data.frame(mean.AUC_cv)%>%
  setNames(shrinkage)%>%
  mutate(n.trees=n.trees)%>%
  pivot_longer(c(-n.trees),names_to = "shrinkage",values_to = "mean.AUC")

df.sd.AUC<-data.frame(sd.AUC_cv)%>%
  setNames(shrinkage)%>%
  mutate(n.trees=n.trees)%>%
  pivot_longer(c(-n.trees),names_to = "shrinkage",values_to = "sd.AUC")

res_cvvv<-df.mean.error%>%mutate(sd.error=df.sd.error$sd.error,
                                 mean.AUC=df.mean.AUC$mean.AUC,
                                 sd.AUC=df.sd.AUC$sd.AUC)

save(res_cvvv, file="../output/res_cvvv.RData")
load("../output/res_cvvv.RData")

if(run.cv){
  p1 <- res_cvvv%>%
    ggplot(aes(x = n.trees, y = mean.error,
               ymin = mean.error - sd.error, ymax = mean.error + sd.error)) + 
    geom_crossbar() +
    facet_wrap(~shrinkage)+
    theme(axis.text.x = element_text(angle = 90, hjust = 1))
  
  p2 <- res_cvvv %>% 
    ggplot(aes(x = as.factor(n.trees), y = mean.AUC,
               ymin = mean.AUC - sd.AUC, ymax = mean.AUC + sd.AUC)) + 
    geom_crossbar() +
    facet_wrap(~shrinkage)+
    theme(axis.text.x = element_text(angle = 90, hjust = 1))
  
  print(p1)
  print(p2)
}

  • Choose the “best” parameter value
n.trees_best <- as.numeric(res_cvvv[which.min(res_cvvv$mean.error),1])
shrinkage_best <- as.numeric(res_cvvv[which.min(res_cvvv$mean.error),2])

n.trees_best
## [1] 50
shrinkage_best
## [1] 0.15
  • Train the model with the entire training set using the selected model (model parameter) via cross-validation. The “best” parameters we got here are n.trees=50 and shrinkage=0.15.
# training weights
weight_train <- rep(NA, length(label_train))
for (v in unique(label_train)){
  weight_train[label_train == v] = 0.5 * length(label_train) / length(label_train[label_train == v])
}

if (sample.reweight){
  tm_train <- system.time(fit_train <- train(feature_train,label_train,
                                             w = weight_train,
                                             n.trees_best,shrinkage_best))
} else {
  tm_train <- system.time(fit_train <- train(feature_train,label_train,
                                             w=NULL,
                                             n.trees_best,shrinkage_best))
}
save(fit_train, file="../output/fit_train.RData")

Step 5: Run test on test images

tm_test = NA
feature_test <- as.matrix(dat_test[, -6007])

if(run.test){
  load(file="../output/fit_train.RData")
  tm_test <- system.time(
    {label_pred <- as.integer(test(fit_train,feature_test,
                                   n.trees_best,shrinkage_best, 
                                   pred.type = 'link'));
    
    prob_pred <- test(fit_train, feature_test, 
                      n.trees_best,shrinkage_best,
                      pred.type = 'response')})
}
## reweight the test data to represent a balanced label distribution

label_test <- as.integer(dat_test$label)

weight_test <- rep(NA, length(label_test))

for (v in unique(label_test)){
  weight_test[label_test == v] = 0.5 * length(label_test) / length(label_test[label_test == v])
}

accu <- mean(label_pred == label_test)
tpr.fpr <- WeightedROC(prob_pred, label_test, weight_test)
auc <- WeightedAUC(tpr.fpr)

cat("The accuracy of model: GBM with n.trees =", n.trees_best, "and shrinkage =",shrinkage_best,"is", accu*100, "%.\n")
## The accuracy of model: GBM with n.trees = 50 and shrinkage = 0.15 is 69.66667 %.
cat("The accuracy of model: GBM with n.trees =", n.trees_best, "and shrinkage =",shrinkage_best,"is", auc, ".\n")
## The accuracy of model: GBM with n.trees = 50 and shrinkage = 0.15 is 0.6835863 .

Summarize Running Time

Prediction performance matters, so does the running times for constructing features and for training the model, especially when the computation resource is limited.

cat("Time for constructing training features=", tm_feature_train[1], "s \n")
## Time for constructing training features= 1.268 s
cat("Time for constructing testing features=", tm_feature_test[1], "s \n")
## Time for constructing testing features= 0.268 s
cat("Time for training model=", tm_train[1], "s \n") 
## Time for training model= 33.51 s
cat("Time for testing model=", tm_test[1], "s \n")
## Time for testing model= 0.487 s

Reference